[SLP] Support unary FNeg vectorization

Differential Revision: https://reviews.llvm.org/D63609

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@364219 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Vectorize/SLPVectorizer.cpp b/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 2da9ead14caaffa0d06aa4ac897a2df2a5d9197d..b3487cb3aa892a7c60513263706690cf84af4304 100644 (file)
--- a/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -2390,6 +2390,7 @@ void BoUpSLP::buildTree_rec(ArrayRef<Value *> VL, unsigned Depth,
       return;
     }
     case Instruction::Select:
+    case Instruction::FNeg:
     case Instruction::Add:
     case Instruction::FAdd:
     case Instruction::Sub:
@@ -2409,7 +2410,7 @@ void BoUpSLP::buildTree_rec(ArrayRef<Value *> VL, unsigned Depth,
     case Instruction::Or:
     case Instruction::Xor: {
       auto *TE = newTreeEntry(VL, true, UserTreeIdx, ReuseShuffleIndicies);
-      LLVM_DEBUG(dbgs() << "SLP: added a vector of bin op.\n");
+      LLVM_DEBUG(dbgs() << "SLP: added a vector of un/bin op.\n");
 
       // Sort operands of the instructions so that each side is more likely to
       // have the same opcode.
@@ -2881,6 +2882,7 @@ int BoUpSLP::getEntryCost(TreeEntry *E) {
       int VecCost = TTI->getCmpSelInstrCost(S.getOpcode(), VecTy, MaskTy, VL0);
       return ReuseShuffleCost + VecCost - ScalarCost;
     }
+    case Instruction::FNeg:
     case Instruction::Add:
     case Instruction::FAdd:
     case Instruction::Sub:
@@ -2918,7 +2920,8 @@ int BoUpSLP::getEntryCost(TreeEntry *E) {
       ConstantInt *CInt0 = nullptr;
       for (unsigned i = 0, e = VL.size(); i < e; ++i) {
         const Instruction *I = cast<Instruction>(VL[i]);
-        ConstantInt *CInt = dyn_cast<ConstantInt>(I->getOperand(1));
+        unsigned OpIdx = isa<BinaryOperator>(I) ? 1 : 0;
+        ConstantInt *CInt = dyn_cast<ConstantInt>(I->getOperand(OpIdx));
         if (!CInt) {
           Op2VK = TargetTransformInfo::OK_AnyValue;
           Op2VP = TargetTransformInfo::OP_None;
@@ -3698,6 +3701,31 @@ Value *BoUpSLP::vectorizeTree(TreeEntry *E) {
       ++NumVectorInstructions;
       return V;
     }
+    case Instruction::FNeg: {
+      setInsertPointAfterBundle(E->Scalars, S);
+
+      Value *Op = vectorizeTree(E->getOperand(0));
+
+      if (E->VectorizedValue) {
+        LLVM_DEBUG(dbgs() << "SLP: Diamond merged for " << *VL0 << ".\n");
+        return E->VectorizedValue;
+      }
+
+      Value *V = Builder.CreateUnOp(
+          static_cast<Instruction::UnaryOps>(S.getOpcode()), Op);
+      propagateIRFlags(V, E->Scalars, VL0);
+      if (auto *I = dyn_cast<Instruction>(V))
+        V = propagateMetadata(I, E->Scalars);
+
+      if (NeedToShuffleReuses) {
+        V = Builder.CreateShuffleVector(V, UndefValue::get(VecTy),
+                                        E->ReuseShuffleIndices, "shuffle");
+      }
+      E->VectorizedValue = V;
+      ++NumVectorInstructions;
+
+      return V;
+    }
     case Instruction::Add:
     case Instruction::FAdd:
     case Instruction::Sub:

diff --git a/test/Transforms/SLPVectorizer/X86/phi3.ll b/test/Transforms/SLPVectorizer/X86/phi3.ll
index 7c8757cd14ca8b0aa8efb5caf4516e7f6954bcee..18653d6cf6d4880e39494c52e7d8985ed32fee53 100644 (file)
--- a/test/Transforms/SLPVectorizer/X86/phi3.ll
+++ b/test/Transforms/SLPVectorizer/X86/phi3.ll
@@ -54,13 +54,11 @@ if.end7:                                          ; preds = %if.then6, %if.then,
 define void @Rf_GReset_unary_fneg() {
 ; CHECK-LABEL: @Rf_GReset_unary_fneg(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[SUB:%.*]] = fneg double undef
 ; CHECK-NEXT:    [[TMP0:%.*]] = load double, double* @d, align 8
-; CHECK-NEXT:    [[SUB1:%.*]] = fneg double [[TMP0]]
+; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> undef, double [[TMP0]], i32 1
+; CHECK-NEXT:    [[TMP2:%.*]] = fneg <2 x double> [[TMP1]]
 ; CHECK-NEXT:    br i1 icmp eq (%struct.GPar.0.16.26* (...)* inttoptr (i64 115 to %struct.GPar.0.16.26* (...)*), %struct.GPar.0.16.26* (...)* @Rf_gpptr), label [[IF_THEN:%.*]], label [[IF_END7:%.*]]
 ; CHECK:       if.then:
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> undef, double [[SUB]], i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x double> [[TMP1]], double [[SUB1]], i32 1
 ; CHECK-NEXT:    [[TMP3:%.*]] = fsub <2 x double> [[TMP2]], undef
 ; CHECK-NEXT:    [[TMP4:%.*]] = fdiv <2 x double> [[TMP3]], undef
 ; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[TMP4]], i32 0

diff --git a/test/Transforms/SLPVectorizer/X86/propagate_ir_flags.ll b/test/Transforms/SLPVectorizer/X86/propagate_ir_flags.ll
index f01b9fe4375f74f9be31cecde32e87ef76d724ab..722d6f8f809ef62029c1f0dfde1347f0d1db293d 100644 (file)
--- a/test/Transforms/SLPVectorizer/X86/propagate_ir_flags.ll
+++ b/test/Transforms/SLPVectorizer/X86/propagate_ir_flags.ll
@@ -559,15 +559,10 @@ define void @fcmp_fast_unary_fneg(double* %x) #1 {
 ; CHECK-NEXT:    [[TMP1:%.*]] = bitcast double* [[IDX1]] to <2 x double>*
 ; CHECK-NEXT:    [[TMP2:%.*]] = load <2 x double>, <2 x double>* [[TMP1]], align 8
 ; CHECK-NEXT:    [[TMP3:%.*]] = fcmp fast oge <2 x double> [[TMP2]], zeroinitializer
-; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x double> [[TMP2]], i32 0
-; CHECK-NEXT:    [[SUB1:%.*]] = fneg fast double [[TMP4]]
-; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[TMP2]], i32 1
-; CHECK-NEXT:    [[SUB2:%.*]] = fneg fast double [[TMP5]]
-; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <2 x double> undef, double [[SUB1]], i32 0
-; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x double> [[TMP6]], double [[SUB2]], i32 1
-; CHECK-NEXT:    [[TMP8:%.*]] = select <2 x i1> [[TMP3]], <2 x double> [[TMP2]], <2 x double> [[TMP7]]
-; CHECK-NEXT:    [[TMP9:%.*]] = bitcast double* [[IDX1]] to <2 x double>*
-; CHECK-NEXT:    store <2 x double> [[TMP8]], <2 x double>* [[TMP9]], align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = fneg fast <2 x double> [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = select <2 x i1> [[TMP3]], <2 x double> [[TMP2]], <2 x double> [[TMP4]]
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast double* [[IDX1]] to <2 x double>*
+; CHECK-NEXT:    store <2 x double> [[TMP5]], <2 x double>* [[TMP6]], align 8
 ; CHECK-NEXT:    ret void
 ;
   %idx1 = getelementptr inbounds double, double* %x, i64 0
@@ -632,15 +627,10 @@ define void @fcmp_no_fast_unary_fneg(double* %x) #1 {
 ; CHECK-NEXT:    [[TMP1:%.*]] = bitcast double* [[IDX1]] to <2 x double>*
 ; CHECK-NEXT:    [[TMP2:%.*]] = load <2 x double>, <2 x double>* [[TMP1]], align 8
 ; CHECK-NEXT:    [[TMP3:%.*]] = fcmp oge <2 x double> [[TMP2]], zeroinitializer
-; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x double> [[TMP2]], i32 0
-; CHECK-NEXT:    [[SUB1:%.*]] = fneg double [[TMP4]]
-; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[TMP2]], i32 1
-; CHECK-NEXT:    [[SUB2:%.*]] = fneg double [[TMP5]]
-; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <2 x double> undef, double [[SUB1]], i32 0
-; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x double> [[TMP6]], double [[SUB2]], i32 1
-; CHECK-NEXT:    [[TMP8:%.*]] = select <2 x i1> [[TMP3]], <2 x double> [[TMP2]], <2 x double> [[TMP7]]
-; CHECK-NEXT:    [[TMP9:%.*]] = bitcast double* [[IDX1]] to <2 x double>*
-; CHECK-NEXT:    store <2 x double> [[TMP8]], <2 x double>* [[TMP9]], align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = fneg <2 x double> [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = select <2 x i1> [[TMP3]], <2 x double> [[TMP2]], <2 x double> [[TMP4]]
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast double* [[IDX1]] to <2 x double>*
+; CHECK-NEXT:    store <2 x double> [[TMP5]], <2 x double>* [[TMP6]], align 8
 ; CHECK-NEXT:    ret void
 ;
   %idx1 = getelementptr inbounds double, double* %x, i64 0